🦅 InsightOS: The Virtual CIO for Real-Time Financial Intelligence

InsightOS transforms complex natural language into actionable executive-level business intelligence. It’s not just an NL2SQL tool; it’s an AI-driven decision engine that reasons across your entire database to provide strategic recommendations.

---

🏗️ System Architecture

Our engine utilizes a Multi-Agent Orchestration Flow powered by LangGraph. This architecture decouples intent classification, schema reasoning, and executive synthesis into a high-precision pipeline.

flowchart TD
%% User Entry
User([Executive Query]) --> API[FastAPI Gateway]
subgraph NL2SQL_Core ["NL2SQL Orchestration Pipeline (LangGraph)"]
direction TB
API --> Intent[Intent & Domain Classifier]
Intent --> Router{Domain Specific?}
Router -->|Operations| D1[Ops Context]
Router -->|Risk| D2[Risk Context]
Router -->|Security| D3[Security Context]
Router -->|Compliance| D4[Compliance Context]
D1 & D2 & D3 & D4 --> Retriever[Few-Shot & Schema Retriever]
Retriever --> SQLGen[SQL Generation Agent]
SQLGen --> Validator{SQL Validator}
Validator -->|Pass| Exec[Secure DB Execution]
Validator -->|Fail| Repair[Self-Repair Agent]
Repair --> SQLGen
end
subgraph CIO_Reasoning ["The Virtual CIO Engine"]
direction TB
Exec --> Viz[Visual Preference Discovery]
Viz --> Insight[Executive Insight Synthesis]
Insight --> Recommendations[Actionable Strategy Generator]
end
%% Result Delivery
Recommendations --> Dashboard([InsightOS Executive Briefing])
Dashboard -.-> |Feedback Loop| User
%% Styling
style User fill:#f9f,stroke:#333,stroke-width:2px
style Intent fill:#6b5b95,stroke:#fff,color:#fff
style SQLGen fill:#f96,stroke:#333
style Validator fill:#feb236,stroke:#333
style Exec fill:#d64161,stroke:#fff,color:#fff
style Dashboard fill:#3ff,stroke:#333,stroke-width:4px
style Recommendations fill:#00d2ff,stroke:#333,stroke-width:2px

Loading

---

🛠️ Pipeline Breakdown

1. Contextual Intent Layer

Unlike generic NL2SQL, InsightOS understands the Domain Context (Risk vs. Ops). This allows the system to differentiate between "flagged" for a fraud reason and "flagged" for a transaction status error.

2. Self-Healing SQL Agent

The system implements a Retry & Repair loop. If the generated SQLite is syntactically correct but functionally fails execution, the error message is fed back to the Agent for an immediate "hot-fix."

3. Executive Synthesis (The CIO)

The final output is not a JSON blob. It is a synthesized report containing:

• The Finding: Direct answer to the query.
• The Insight: Root cause analysis or trend detection.
• The Recommendation: A suggested business action (e.g., "Adjust credit limits for High-Risk users").

---

✨ Key Features

• 🛡️ Multi-Domain Intelligence: Dedicated prompts for Operations, Risk, Compliance, and Security.
• 📊 Dynamic Visualization: Automatically chooses the best chart (Bar, Line, Pie, or Table) based on data distribution.
• ⚡ Sub-Second Latency: Optimized query generation using Gemini 2.5 Flash.
• 🔌 Plug-and-Play: Seamless integration with existing SQLite/SQLAlchemy databases.

---

🚀 Getting Started

1. Environment Setup

python -m venv venv
./venv/Scripts/activate # Windows
pip install -r requirements.txt

2. Configuration

Create a .env file with your Google Gemini API Key:

GOOGLE_API_KEY=your_gemini_keyDB_PATH=./derivinsightnew.db

3. Run the Platform

python -m uvicorn app.main:app --reload --port 8080

---

💎 Demo Verification (Try These)

Domain	Executive Question	Impact
Risk	"Show me all users and their risk levels."	Critical Profile Alert
Security	"Show me failed logins by reason and IP address."	Threat Intelligence
Growth	"Which countries have the highest active users?"	Market Optimization
Fraud	"List all transactions in 'FLAGGED' status."	AML Monitoring

---

Developed for the Deriv Hackathon – Empowering Executives with Data-First Decisioning.